Social Signals
AI in the wild: Confident, wrong, and weirdly expensive
Jun
Think about working in search engine marketing for years, researching an issue you understand inside and outside, solely to have an LLM confidently clarify why your expertise is mistaken.
That occurred to me. Really, it occurred three totally different instances final week with Gemini.
The scary half wasn’t that the solutions have been clearly dangerous. It was that they sounded good. The responses have been polished, plausible, and directionally correct sufficient that most individuals would by no means query them.
And for those who aren’t deeply acquainted with the subject, you in all probability wouldn’t know the right way to problem the reply.
Humorous sufficient, I caught it twice. The third time price me cash. All throughout the similar week.
Instance 1: Gemini educates me on technical search engine marketing
Let me provide you with some context.
I’m at present serving to migrate a shopper’s FAQ hub from a provider-hosted subdomain to a self-hosted implementation.
The FAQ lives underneath a /faq/ folder, however particular person Q&A pages are parameter-based URLs. Usually not a problem… besides Shopify forces canonicals again to the foundation /faq/ web page, successfully stopping these pages from being listed.
Whereas researching Shopify-specific options and duplication concerns, Gemini gave me this response:
Excuse me?!
You’ll completely not be penalized for conflicting search engine marketing indicators.
At greatest, Google indexes what it desires. At worst, it ignores directives it doesn’t belief.
The larger problem is the wording.
“Penalty” is mainly the magic phrase in search engine marketing. The second management hears it, priorities shift, momentum dies, and suggestions develop into tougher to implement.
Then Gemini doubled down.
I requested whether or not eradicating the canonicals and permitting the parameter pages to exist independently was an choice.
Gemini responded:
- “Google typically ignores question parameters.”
Wow… simply wow.
Beneath is an implementation I labored on with the Saatva workforce, the place we deliberately listed parameter URLs throughout the procuring expertise.
Search Console and URL inspection confirmed these pages listed simply wonderful.
Parameter pages can completely rank, index, and generate worth.
The issue isn’t that Gemini was mistaken (as in it’s a harder implementation).
It’s that the reply sounded plausible sufficient that somebody with out search engine marketing expertise would in all probability settle for it and transfer on.
Instance 2: Gemini says clear up the difficulty with a $3,000 half
This one hit otherwise as a result of I’m not a mechanic.
I’ve been troubleshooting a problem on my Jeep SRT and spent hours outdoors within the solar accumulating information, testing fuses, reviewing OBD2 logs, and attempting to slender down the foundation trigger. After reviewing every part, Gemini confidently concluded the difficulty was seemingly a rear differential failure and recommended a full alternative.
Estimated injury? Roughly $3,000 in OEM elements alone.
The response sounded incredible. It was detailed, logical, and even complimented my troubleshooting course of.
The issue? It was mistaken. Actually mistaken.
After pushing again and sharing further OBD2 information I had been monitoring throughout analysis, Gemini fully reversed course and admitted it jumped to a worst-case state of affairs with out sufficient proof.
In different phrases, I virtually spent 1000’s changing elements I didn’t want.
That’s what makes these examples fascinating. In search engine marketing, I instantly knew Gemini was mistaken as a result of I had years of expertise to problem the advice. Right here, I didn’t have that benefit. I needed to depend on skepticism, proceed testing, and keep away from treating the reply as reality.
Similar AI. Similar confidence. Utterly totally different consequence.
Instance 3: Gemini price me $20 million {dollars}!
OK, technically, this was online game cash.
I used to be enjoying Madden, working by means of workforce logistics, and attempting to optimize wage cap spending to re-sign key gamers and preserve the roster collectively.
Sooner or later, I obtained lazy and tagged in my associate, Gemini. I shared a screenshot of the workforce funds and requested for a roadmap to restructure contracts and optimize the cap state of affairs.
Gemini gave me an in depth plan. I adopted it.
And it put me $20 million over the wage cap.
The humorous half is that this completely mirrors the sooner examples.
The reply appeared good. It was organized, assured, and gave me a transparent player-by-player motion plan.
So I trusted it. Then I referred to as Gemini out.
You’ll discover within the second screenshot that Gemini basically factors out that I blindly trusted the recommendation with out validating the maths.
Truthfully… truthful.
The distinction is that this error solely price me faux cash and a Madden franchise. The Jeep instance virtually price me actual cash, and the search engine marketing instance may have price implementation momentum and belief.
The worth of experience was by no means memorizing solutions. It was figuring out when one thing feels off, asking higher questions, and recognizing when the reply smells like bullshit.
AI didn’t take away that talent. If something, it made it extra useful.
AI isn’t changing specialists. It’s changing individuals who have stopped considering.
#wild #Assured #mistaken #weirdly #costly
